Hackathon Ideas/MCP AI

[stone veldt]

TLDR: Planning a hackathon around MCP with a twist—want to explore MCP + agentic workflows via Dagger, ideally running as a traditional MCP server with HTTP-streamed output so clients like GitHub Copilot can call it. Dagger offers multi-language flexibility (Go, TS, Java), but it’s complex and slower to ramp up. Need feedback on feasibility and logging for compliance.

⸻

Next week, we’re running a hackathon focused on Model Context Protocol (MCP). I’m debating between two routes:
• Building MCP functions directly with the default MCP SDK
• Using Dagger to construct MCP-related workflows

Here’s the catch:

This isn’t really MCP vs Dagger. I want to explore MCP and agentic workflow via Dagger but not certain if we can make it run as a traditional MCP server-meaning it streams output over HTTP or SSE in a way clients like GitHub Copilot can consume. The goal is not just terminal usage but having a proper client invoke it. I’m unclear if that’s doable outside the Dagger Shell or if it requires shell-bound execution.

The downside: Dagger introduces complexity such as containers, shell abstractions that may slow down beginners or interns, especially in a tight 2-day hackathon window.

Company’s primary language is C#, but for this hackathon, we want to lean into Go, TypeScript, and Java.

Also can anyone let me know if dagger supports local web traces now or still requires dagger cloud login? Asking due to compliance concerns at work which might restrict the cloud version and pure cli debugging might be tough for folks newer to all this.

I want to hit two birds with one stone: push Dagger exploration and meet MCP goals, including agentic workflows. Is this doable in a hackathon or would you advise standard mcp SDKs for this scope

Yes I used gpt to help with post. 🤫
Helped keep it on target and organized.

I'm not a bot

[stone veldt]

<@&946480760016207902> only mentioning once but really looking for your advice if seems like this would be a good fit considering the time crunch or might be a less ideal situation? Could help raise more awareness internally but haven't explored llm (azure OpenAI) with dagger and afraid things might be too much for folks in short hackathon.

Sorry for the direct mention but under a time crunch to decide in next 2 days

Note the company is adopting GitHub copilot so pure cli usage would be a hard sell. I'd need to know if mcp and copilot usage would be possible for demos as this would help as well.

[worldly wadi]

hi @stone veldt !! Thanks for asking for advice here. We've been running 4-hour hack nights, so we've learned a few things along the way.

Here's the pre-recs that we ask everyone to come to the hack night with:
​- A laptop
​- A working container runtime (Docker). Docker must be installed and running before you arrive.
​- Your IDE of choice (VS Code, Zed, Cursor, etc.)

To kick off the hack night, we start with a 15 min intro to quickly explain what Dagger is, share a few examples, and then dive into the set-up together. We've learned that if we don't start with the set up then we lose people. Feel free to use these slides: https://docs.google.com/presentation/d/1J8-r3hR_D0WliedmHAbuxfmf8wwhH1BOIPzT6KhiEwM/edit?slide=id.g3528cc9775e_7_42#slide=id.g3528cc9775e_7_42

We could use the https://discord.com/channels/707636530424053791/1361948780575854612 channel as a place for people to ask questions during your hackathon too! If you let us know the hours, we should be able to help support during the event.

I will let @wise spear @tardy dune @peak marsh give the cloud and GitHub copilot advice

[stone veldt]

Bump. Making call today so knowing if fits the mcp goals would help me confirm if fits or might be too much for the time frame and knowledge level of the folks joining in currently. They are sharp but maybe dagger will add too much to the mix. Any chance someone gets to provide some guidance on this is appreciated.

[brittle bolt]

i think we probably need to know more about your participants... are they platform engineers? or webapp developers? how many are there?

not certain if we can make it run as a traditional MCP server-meaning it streams output over HTTP or SSE in a way clients like GitHub Copilot can consume
currently dagger mcp only has stdio transport support, so that might be a problem if y'all can only run in the cloud and can't run stdio servers on the same host as your agent/MCP client process. also fwiw a "traditional" mcp server is not sse, but stdio... it's a short tradition, though 😂

dagger mcp is also very experimental. it can be used to expose modules to an agent running on the host, but if you've not played with it yet yourself, it may be difficult to sheperd a bunch of hackathon participants in that direction, especially if they're also busy learning dagger in languages they don't write every day.

[brittle bolt]

(also, just feedback for your bot-aided post: i feel like have to read between the lines to make a determination as to whether you're organizing a hackathon for others to participate in or planning a hackathon project for yourself. --- "shell abstractions that may slow down beginners or interns, especially in a tight 2-day hackathon window." is the main thing that allows me to differentiate, but because a bot was involved in the writing i can't be confident i'm inferring from what you wrote or what the bot may've mis-inferred from your prompting)

[brittle bolt]

If you’re actually asking if you should use dagger mcp for your personal or group project in an mcp hackathon, the answer is an am emphatic yes

[wise spear]

@stone veldt GH Copilot in VS Code?

[stone veldt]

• The audience is intern/junior engineers who are interested in pairing with me and working on some functionality to do with MCP.
• We use github copilot and have MCP enabled in it
• My main options here are 1) use dagger to create MCP servers that copilot could call and I can run via shell 2) stick with mcp sdk's and focus on that for now

I like the idea of diving into dagger more, as I've not had time lately for anything interesting, but if the process to work through issues is trickier with dagger involved then maybe it's not a good fit for a 2 day hackathon with folks not really used to what dagger is solving.

And while I don't want to be overly careful on learning curve, realistically I'll set myself and them up for a bad 2 day hackathon if it's adding a lot of extra complexity over just a simple local server running.

My experience: I piloted a MCP c# server internally and had it responding with api calls to an internal platform. used aspire to run/coordinate.

Their experience level is new to all of it. So I'm trying to get the advice if the dagger side might be better saved for my own hackathon in future, or you think it's a viable dev route for newer folks too (I only know of the TUI and limited web trace exposure, and it's not easy stuff for first timers... maybe this has changed)

thank you for the help!

[stone veldt]

i don't use gpt to cleanup my text much but was on mobile. It was me being unclear not the summary. It's for me but I'm also trying to include some folks that are new/interns so you got it right either way.

My main issue was figuring out if it supported operating as mcp server so i could test with github copilot.

I'll look at mcp docs/info, but I think I'm gonna save it for me to do my own hacking session, and not try to teach dagger + mcp at the same time 🙂

I think letting them pick a stack and just go for it with coding agent + default sdk would be a good start and I'll have fun with dagger myself soon 😉

[wise spear]

Sounds good!
If you can use MCP / Copilot / VS Code like https://docs.github.com/en/copilot/customizing-copilot/extending-copilot-chat-with-mcp?tool=vscode

Then you could also throw in https://github.com/dagger/container-use as a MCP server. It needs Dagger to be installed, but you don't need to learn Dagger. It creates containerized environments for coding agents to work locally without trampling on each other like Solomon showed in his keynote at the AI Engineer World's Fair 🙂
https://www.youtube.com/watch?v=U-fMsbY-kHY&t=3400s

[stone veldt]

Im on MCP track recording and didn't know Solomon had done a talk. I don't get as much time as I'd like to explore dagger so will check that out asap!

I have a new dagger project I'm trying to wrap a cli from and honestly just slow for me figuring out the abstractions with pure Go for the git tagging and such. Getting close though. Just need a solid session or two and I'll grok it. Appreciate it!

[stone veldt]

Nice! Anyone built mcp server itself to respond from dagger using go tool mcp https://pkg.go.dev/golang.org/x/tools/internal/mcp

? would love to watch some videos and check out your code

The demo of container use was great. I love how informal and engineers focused it was. Live demos with things is scary sometimes but more entertaining and feels authentic

[wise spear]

One of our devs just saw that package this morning! Exciting that it could help add MCP client support to Dagger.

[brittle bolt]

(The dev was me lol)

I’m stoked, last time I tried to make the engine an mcp client the available libraries fought me real hard

[stone veldt]

I was going to try and add mcp client support for my slackbot. It uses azure OpenAI. I found mcp-go has in memory mcp stuff and more so figured I'd explore. I want to deploy some mcp containers in the same mcp container group and try to leverage with my bot for basic fetch, medium research, context7 docs.

Anyone have luck with mcp client and Go yet? @brittle bolt ?

I'm not looking for fancy just thinking of trying to make my bot more agentic here as a first step.

[brittle bolt]

Haven’t tried yet, lmk how it goes. I heard a rumor last week they’re announcing it soon.

[stone veldt]

Who? Go?
I didn't see updates for a bit so figured they'd be slow to adapt to the spec and might be better to use mcp-go for a while.

[brittle bolt]

Yeah the go maintainers and that package you linked above